doi:10.3926/jiem.2009.v2n2.p360-386 ©© JIEM, 2009 – 2(2): 360-386 - ISSN: 2013-0953
Vehicle-based interactive management with multi-agent approach 360
Y. M. Chen; B.-Y. Wang
Vehicle-based interactive management with multi-agent
approach
Yee Ming Chen, Bo-Yuan Wang
Department of Industrial Engineering and Management, Yuan Ze University
(TAIWAN, ROC)
[email protected]; [email protected]
Received May 2009 Accepted September 2009
Abstract: Under the energy crisis and global warming, mass transportation becomes more
important than before. The disadvantages of mass transportation, plus the high flexibility
and efficiency of taxi and with the revolution of technology, electric-taxi is the better
transportation choice for metropolis. On the other hand, among the many taxi service
types, dial-a-ride (DAR) service system is the better way for passenger and taxi. However
the electricity replenishing of electric-taxi is the biggest shortage and constraint for DAR
operation system. In order to more effectively manage the electric-taxi DAR operation
system and the lots of disadvantages of physical system and observe the behaviors and
interactions of simulation system, multi-agent simulation technique is the most suitable
simulation technique. Finally, we use virtual data as the input of simulation system and
analyze the simulation result. We successfully obtain two performance measures: average
waiting time and service rate. Result shows the average waiting time is only 3.93 seconds
and the service rate (total transport passenger number / total passenger number) is
37.073%. So these two performance measures can support us to make management
decisions. The multiagent oriented model put forward in this article is the subject of an
application intended in the long term to supervise the user information system of an urban
transport network.
Keywords: multiagent, dispatch, dial-a-ride problem
brought to you by COREView metadata, citation and similar papers at core.ac.uk
provided by UPCommons. Portal del coneixement obert de la UPC
doi:10.3926/jiem.2009.v2n2.p360-386 ©© JIEM, 2009 – 2(2): 360-386 - ISSN: 2013-0953
Vehicle-based interactive management with multi-agent approach 361
Y. M. Chen; B.-Y. Wang
1 Introduction
To resist global warming, mass transportation plays an important role in metropolis
(Government of Hong Kong, 2005). Public can travel by taking mass transportation
rather than driving a car. In this way, public can reduce the CO2 generation and air
pollution (GreenPartyTaiwan, 2007). In order to reduce the operation cost of mass
transportation, Transportation Company has to abandon the flexibility and
efficiency (Wu, 2006). However taxi just compensates the disadvantages of mass
transportation, because it has the features of high flexibility and efficiency
(GreenPartyTaiwan, 2007). So far, most of passengers stand beside road and wait
for a taxi. In this condition, there are three disadvantages. First, passenger doesn’t
know how long he/she has to wait until a taxi passes by. Secondly, passenger is
not sure the coming taxi is free or not. Thirdly, taxi needs to go around and look
for passenger. Hence the operation utility is too low so that causes the energy
waste (Wu, 2005). Contrary, dial-a-ride (DAR) system is a good solution to solve
this problem. In DAR system, passengers use wireless communication tool (mobile
phone) to call for a pick-up and delivery service-to-service center (control center).
Then, service center assigns an idle taxi to perform the task. Using this kind of
service system, passenger doesn’t need to wait longer than before. And taxi driver
can save the taxi energy. Hence, Dial-a-ride system is very important in taxi
operation system (Wu, 2005). On the other hand, with the technological revolution
of power, electric-taxi comes with the tide of fashion (Taiwan Environmental
Information Center, 2009; BigSolar, 2005). Electric taxi means a taxi is driven by
electricity. It has two advantages. First, for the earth, electric taxi can reduce the
air pollution and global warming, because it can’t emit CO2 (the electric taxi we
talk about is driven by pure electricity. The pure electricity means that it doesn’t
emit CO2 and its source doesn’t involve any organic compound of carbon, like
hydrogen-battery. For hydrogen-battery itself, its waste is “water” so it indeed can
reduce the emission of CO2. (iCo2l, 2008)). Secondly, for taxi driver, electric-taxi
can reduce the fuel cost, especially for oil, due to the cost of replenishing electricity
is lower than gas or gasoline. Hence electric taxi is an important transportation for
metropolises.
Contrary, electric-taxi also has disadvantage. The worst disadvantage in electric-
taxi is the electricity that’s also the biggest limitation (Galus et al., 2009). During
the electric-taxi traveling period, the electricity of taxi is decreasing. When the
electricity of electric-taxi is not enough to do the next service, electric-taxi has to
doi:10.3926/jiem.2009.v2n2.p360-386 ©© JIEM, 2009 – 2(2): 360-386 - ISSN: 2013-0953
Vehicle-based interactive management with multi-agent approach 362
Y. M. Chen; B.-Y. Wang
replenish its’ electricity in the electric station. During the replenishing period,
electric-taxi can’t do any task and passenger still waits for service. Under this
condition, that will cause the reduction of taxi company’s revenue and passenger
satisfaction (GreenPartyTaiwan, 2007). However, revenue and passenger
satisfaction are the most important performance measures for Taxi Company. So
Electric-taxi Company has to propose some management policies to deal with the
electric replenishing problem (Galus et al., 2009). Hence how to manage the
electric-taxi DAR operation system becomes a very important problem with
management policies (GreenPartyTaiwan, 2007; Taiwan Environmental Information
Center, 2009).
In order to manage the electric-taxi DAR operation system, we have to construct
an electric-taxi DAR operation simulation system. There is a lack in multi-agent
transportation simulation, such as allowing cars move based on shortest path and
dispatching operations. In fact, the traffic jams management is considerable for
electric-taxi DAR operation system (Ezzedine et al., 2005; Kok & Lucassen, 2007;
Lansdowne, 2006; Cubillos et al., 2008). So this paper takes into account the
shortages of existing methods to reinforce our multi-agent simulation. On the other
hand, due to the impracticable and costly weaknesses of physical system (Ali,
2006), multi-agent simulation technique is the most suitable simulation technique
for our research.
The main purposes of this study are as follows: First purpose is to provide a series
of management policies to manage the electric-taxi DAR operation system and
analyze the phenomenon of simulation. Second purpose is to compensate the
shortages of existing methods to reinforce our multi-agent simulation. The main
contribution of this paper is that we successfully obtain the performance measures
(average waiting time and service rate) to support the decision making for
manager.
The rest of this paper is described as follows. Section 2 is the literature review.
Section 3 introduces the electric-taxi DAR operation system. Section 4 creates the
simulation system and describes the environment setting. Section 5 is to collect
and analyze the data obtained from simulation. The last section will make a
conclusion that includes the contributions of this research and describe the future
work.
doi:10.3926/jiem.2009.v2n2.p360-386 ©© JIEM, 2009 – 2(2): 360-386 - ISSN: 2013-0953
Vehicle-based interactive management with multi-agent approach 363
Y. M. Chen; B.-Y. Wang
2 Literature review
Characteristics
Ezz
edin
e et
al.
(2005)
Lansd
ow
ne
(2006)
Kok
& L
uca
ssen
(2
007)
Cubill
os
et a
l.
(2008)
Seo
w e
t al
. (2
008)
Gal
us
et a
l. (
2009)
Our
pro
pose
d
met
hod
Being generic * * * * * * * Being agent-based simulation
* * * * * * *
Multiple (people) to multiple (cars) simulation
* * * * * * *
Taking into account the energy wasting of cars
* *
Taking into account the change of cars’ velocity
* *
Information management decentralization
* * * * *
Hybrid information management (Combining centralization and decentralization)
* *
Taking into account traffic jam
* * *
The traffic jam management (such as allowing cars move that excludes the road which is under traffic jam)
* *
Allowing cars move based on shortest path
* * *
Table 1. “Criteria distinguishing our method from those existing in literature review”.
According to the literature review, this paper lists ten main characteristics (see
Table 1). And these characteristics are considerable for vehicle DAR operation
system (Cubillos et al., 2008). Moreover, we distinguish our method from those
existing methods in literature review. Obviously, these existing methods are
incomplete in vehicle DAR operation system. So these shortages can be the
improvements of our electric-taxi DAR operation simulation.
3 Electric-taxi DAR operation system
In this chapter, we describe the electric-taxi DAR operation system (Figure 1). We
divide the electric-taxi DAR operation system into three parts. First part is the
agent container. In agent container, we have six kinds of agents: electric-taxis,
control center, passengers, electric stations, road and stops. Agents are divided
into two types (Figure 2). First type is active agents those have their own
doi:10.3926/jiem.2009.v2n2.p360-386 ©© JIEM, 2009 – 2(2): 360-386 - ISSN: 2013-0953
Vehicle-based interactive management with multi-agent approach 364
Y. M. Chen; B.-Y. Wang
structures and behaviors. Second type is passive agents those only have their
structures.
Figure 1. “The framework of electric-taxi DAR operation system”.
The specification of each type of agent is as follows:
Passive Agents (PA): This category of agents represents entities (agents)
that have a structure, without a behavior. Usually, a large part of the
elements contained in the simulation environment belongs to this category.
For example, road is a passive agent, its x and y coordinates belong to
spatial structure. And, its id and color belong to non-spatial structure.
Active agents (AA): This category of agents represents entities (agents),
which have structures and behaviors: These entities actively participate in
the simulation. For this category, we must specify the data structures of the
entities (spatial and non-spatial structures) as well as their behaviors
doi:10.3926/jiem.2009.v2n2.p360-386 ©© JIEM, 2009 – 2(2): 360-386 - ISSN: 2013-0953
Vehicle-based interactive management with multi-agent approach 365
Y. M. Chen; B.-Y. Wang
(spatial and non-spatial behaviors). For example, electric taxi is an active
agent. Its x and y coordinates belong to spatial structure. Its id and color
belong to non-spatial structure. Its moving behavior belongs to spatial
behavior. Its information transmitting belongs to non-spatial behavior.
Figure 2. “Passive Agents and Active agents”.
The second part is environment. Environment is composed of time container, data
container, event container and scenario container. Time container includes all of
the related information and variables. Data container contains all of the related
information and variables to support data analysis. Event container has many
events to be the triggers of every possible action of each agent. We also design
three kinds of scenarios in the scenario container (Figure 1): Incident management
(traffic jam), dial-a-ride management and electricity replenishing management.
These three kinds of scenarios constitute the underlying scenario of electric taxi
DAR operation system. The details of these three kinds of scenarios will be
specified in sub-section 3.2. In our data analysis, we want to know how the
performance (average waiting time) is via scenario simulation. And then we will
make a recommendation and conclusion.
Third part, we design management policies to manage the electric-taxi DAR
operation simulation system in the Section 4. In this section, we will specify the
framework of active agents which belong to agent container and the scenario
interaction processes of scenario container in the environment.
doi:10.3926/jiem.2009.v2n2.p360-386 ©© JIEM, 2009 – 2(2): 360-386 - ISSN: 2013-0953
Vehicle-based interactive management with multi-agent approach 366
Y. M. Chen; B.-Y. Wang
3.1 The framework of active agents
Following, we will describe the framework of active agent and define each function
within the framework of active agent. Through this section, we can construct the
framework of each kind of active agent clearly. Although we can’t clearly realize
the exact actions, attributes and duties what the agents will have through this
framework immediately. However it can convert the abstract structure of each
active agent the active agent into a concrete framework. And in this way, we can
extract the actions, attributes and duties what the agents will have through this
framework further.
The functions within the framework of active agent:
K: Knowledge base of agent. The knowledge base of agent can help agent
make decision more precisely. Some agents have their own knowledge
base. Knowledge base consists of a lot of information. Information comes
from different kinds of agents. For example, the information of control
center comes from passenger and electric taxi. This information makes up
the knowledge base of control center. The difference of knowledge base
comes from roles (electric taxi, passenger or control center) not types
(passive agent or active agent). Different role may have different
knowledge base.
IS: Information source. Information source means where the information
comes from.
R: Response. Response means the object of action.
D: Decision rules. Decision means the alternatives, mechanisms or
algorithm to help agent make proper actions or decisions.
In this section, we illustrate six kinds of agents: CCA Control Center agent; PA:
Passenger agents; TA: Electric Taxi agents; EA: Electric station agents; RA: Road
agents; and SA: Stop agents.
doi:10.3926/jiem.2009.v2n2.p360-386 ©© JIEM, 2009 – 2(2): 360-386 - ISSN: 2013-0953
Vehicle-based interactive management with multi-agent approach 367
Y. M. Chen; B.-Y. Wang
Figure 3. “The framework of Control Center agent”.
The first and foremost, we specify the framework of active agent through CCA. In
figure 3, for perception, we use IS-TA to represent that Control Center agent
perceives the information (IS) where comes from Electric-taxi agents (TA). For
knowledge base, we use K-TA to represent that Control Center agent owns the
knowledge (K) about Electric-taxi agents (TA). For decision rules, we use D-T to
represent that Control Center agent owns the decision rule (D) about Electric-taxi
agents (T). For action, we use R-TA to represent that Control Center agent
responses (R) to Electric-taxi agents (TA). The rest can be deduced by analogy.
Following, we specify the framework of each type of agent.
In figure 3, there are three types of information source: first type is simulation
environment. Second type is passenger agents. And third type is electric-taxi
agents. For CCA, it has four kinds of knowledge: (1) the information about Electric-
taxi agents; (2) road and electric station; (3) display parameters and (4) CCAs’
attributes. Combine the knowledge and information that perceived, and as the
input data of decisions. There are four decision rules: (1) passenger service rule;
(2) taxi chosen rule; (3) stop chosen rule and (4) electric station chosen rule.
Through decision rules, CCA executes proper actions to two kinds of agents:
Electric-taxi agents and Passenger agents. The final action is that the behavior of
CCA will be displayed on the screen via display parameters.
doi:10.3926/jiem.2009.v2n2.p360-386 ©© JIEM, 2009 – 2(2): 360-386 - ISSN: 2013-0953
Vehicle-based interactive management with multi-agent approach 368
Y. M. Chen; B.-Y. Wang
For passenger agents, Electric-taxi agents and Electric station agents, we just
describe the difference of PA, TA and EA respectively. Next, we describe the
difference of framework of PA to Control Center agent. In figure 4, for perception,
PA doesn’t include the information source of PA but PA contains the information
source of Control Center agent. For knowledge base, PA doesn’t include the
knowledge base of control center, electric-taxi, road and electric station but PA
contains the knowledge base of its’ attributes. For decision rules, PA only has a
decision rule of calling. For action, PA doesn’t include the response to the
Passenger agents but PA contains the response to the Control Center agent.
Figure 4. “The framework of Passenger agent”.
We also describe the difference of framework of TA to Control Center agent. In
figure 5, for perception, TA contains the information source of Control Center
agent. For knowledge base, TA doesn’t include the knowledge base of control
center but TA contains the knowledge base of its’ attributes. For decision rules, TA
contains shortest path searching rule, avoid traffic jam rule, electric station chosen
rule, replenishing electric rule and stop chosen rule. And only the stop chosen rule
is repeated with control center. For action, TA contains the response to the Control
Center agent.
We describe the difference of framework of EA to Control Center agent. In figure 6,
for perception, EA only contains the information source of electric-taxi . For
knowledge base, EA only contains the knowledge base of its’ attributes and display
doi:10.3926/jiem.2009.v2n2.p360-386 ©© JIEM, 2009 – 2(2): 360-386 - ISSN: 2013-0953
Vehicle-based interactive management with multi-agent approach 369
Y. M. Chen; B.-Y. Wang
parameters. For decision rules, EA only contains service rule. For action, EA only
contains the response to the electric-taxi agent and displayed on the screen via
display parameters.
Figure 5. “The framework of Electric-taxi agent”.
Figure 6. “The framework of Electric station agent”.
Next, we describe the difference of framework of RA to Control Center agent. In
figure 7, for perception, RA only contains the information sources of electric-taxi
doi:10.3926/jiem.2009.v2n2.p360-386 ©© JIEM, 2009 – 2(2): 360-386 - ISSN: 2013-0953
Vehicle-based interactive management with multi-agent approach 370
Y. M. Chen; B.-Y. Wang
and simulation environment. For knowledge base, RA only contains the knowledge
base of display parameters. For decision rules, RA only contains road state change
rule. For action, RA only contains the response of displaying on the screen via
display parameters.
Figure 7. “The framework of Road agent”.
Figure 8. “The framework of Stop agent”.
Finally, we describe the difference of framework of SA to Control Center agent. In
figure 8, for perception, SA only contains the information source of simulation
doi:10.3926/jiem.2009.v2n2.p360-386 ©© JIEM, 2009 – 2(2): 360-386 - ISSN: 2013-0953
Vehicle-based interactive management with multi-agent approach 371
Y. M. Chen; B.-Y. Wang
environment. For knowledge base, SA only contains the knowledge base of display
parameters. For decision rules, SA only contains stop position transferring rule. For
action, SA only contains the response of displaying on the screen via display
parameters.
3.2 The interaction processes diagram of scenario
As mentioned above, scenario container has three kinds of scenarios. In this sub-
section, we specify the interaction process of each scenario. The three scenarios
interaction processes are: (1) dial-a-ride scenario; (2) incident (traffic jam)
scenario; and (3) electricity replenishing scenario. The detail of each scenario is
described as follows.
Figure 7 is the dial-a-ride interaction process diagram. We draw an interaction
process diagram to help us to concrete the abstractive interaction concept. And we
can realize the sequence of interaction between agents so that we can easily
construct the scenario with program language in simulation platform. In the
scenario interaction processes diagram, a rectangle represents an active agent that
also means the executor of an event/behavior. The dotted line represents the state
of actor/agent is static, which means actor/agent doesn’t do any action. The
straight thick line represents the state of actor is dynamic with time. The length of
straight thick line represents how long the behavior/event will last. The transverse
thick line represents an event/behavior is executed. The arrow of transverse thick
line represents the interaction object of an event/behavior execution. Finally, the
number, which is in front of every event/behavior, represents the execution
sequence. Following, we specify the scenario and management problems of each
scenario. The management policies of management problems for each scenario will
be specified in the 3.3 sub-section.
Dial-a-ride scenario
The dial-a-ride scenario (figure 8): Passenger calls for a service to control center.
After control center receiving a requirement, control center assign a free electric-
taxi to pick the specific passenger up and deliver him/her to the destination. The
main roles are passenger, control center and electric-taxi.
The management problems of dial-a-ride scenario: (1) How to select the passenger
who you want to pick-up. (2) How to select a proper taxi to perform the service
doi:10.3926/jiem.2009.v2n2.p360-386 ©© JIEM, 2009 – 2(2): 360-386 - ISSN: 2013-0953
Vehicle-based interactive management with multi-agent approach 372
Y. M. Chen; B.-Y. Wang
task. (3) How to choose the path to move to passenger location/destination. (4)
What is the next step to driver, after driver delivering passenger to his/her
destination.
Figure 9. “Dial-a-ride interaction process”.
In dial-a-ride interaction process diagram (figure 9), there are three agents:
Control Center, Electric-taxis and Passengers. The detail of process is described as
below: In the beginning user execute the simulation (step1). Immediately,
simulation initial and generate three kind of agent (step2-1, 2-3, 2-4 and 2-5), and
simulation time is begin running and updating (step2-2). After generating, Control
Center and Electric-taxi stand by for Passenger demand (step3-1 and 3-2). On the
other hand, passenger sends a requirement to control center and waiting for
response (step3-3). Control Center assigns a most suitable Electric-taxi for
Passenger and stand by for next demand (step4, 5-1 and 5-2). Then assigned-
electric-taxi finds the shortest path to move to pick passenger up (step6, 7 and 8).
After arriving at Passenger location, Electric-taxi delivers Passenger to his
destination (step9-1 and 9-2). When traveling finished, Electric-taxi stand by in the
current position as well as Passenger gets off and leave the system (step10-1 and
10-2).
doi:10.3926/jiem.2009.v2n2.p360-386 ©© JIEM, 2009 – 2(2): 360-386 - ISSN: 2013-0953
Vehicle-based interactive management with multi-agent approach 373
Y. M. Chen; B.-Y. Wang
Incident (traffic jam) scenario
The incident (traffic jam) scenario (figure 10): An Electric-taxi encounter a road
which is crowded, the Electric-taxi finds out another shortest path that excludes
the road which is crowded to avoid the traffic jam. Any Electric-taxi which is getting
in the traffic jam will send the traffic jam information to each Electric-taxi. The
main role is Electric-taxis.
The management problems of incident (traffic jam) scenario: (1) How to know the
specific road, which is under traffic jam in any time. (2) How to avoid the specific
road, which is under traffic jam keep and moving on.
Figure 10. “Incident (traffic jam) interaction process”.
In this incident (traffic jam) interaction process (figure 10), there is one type of
agents: Electric-taxis. The detail of process is described as below: In the beginning
user execute the simulation (step1). Immediately, simulation initial and generate
two kinds of actors (step2-2 and 2-3), and simulation time is begin running and
updating (step2-1). During the simulation period, road will update its’ state (traffic
jam/ freely flowing) (step3). After generating and simulating for a while, Electric-
taxi can be divided into two kinds of Electric-taxis. Some Electric-taxis are getting
in traffic jam and others are moving freely (represented by the longest transverse
thick-line). The Electric-taxis those are getting in traffic jam send traffic jam
information to each other and other Electric-taxis those are not getting traffic jam
doi:10.3926/jiem.2009.v2n2.p360-386 ©© JIEM, 2009 – 2(2): 360-386 - ISSN: 2013-0953
Vehicle-based interactive management with multi-agent approach 374
Y. M. Chen; B.-Y. Wang
(step4-2, 4-3 and 4-4). When Electric-taxis enter in or leave road, the state of road
will be changed into jam-packed (step4-1 and 6-1). When Electric-taxi moves to a
traffic jam road, driver will find another shortest path that excluded the traffic jam
road (step5 and 6-2). And then Electric-taxi will switch into another road and move
on (step7). Finally, when Electric-taxis that are leaving the traffic jam road, they
will sending the traffic jam information again (step8).
Electricity replenishing scenario
The electricity replenishing scenario (figure 11): An Electric-taxi driver checks the
electricity of taxi, and find out the electricity is too low to support any longer
travel. So the Electric-taxi moves to electric station to replenish electricity. Control
Center and Electric-taxis those are in the Electric stations will provide the electric
station information to the electric-taxi which needs to replenish electricity. The
main roles are: Electric-stations, Control Center and Electric-taxis.
The management problems of electricity replenishing scenario: (1) How to
schedule Electric-taxis those are waiting for replenishing. (2) When an Electric-taxi
should be replenished. (3) How long should an Electric-taxi replenish electricity?
(4) How to select a proper Electric station. (5) After finishing the replenishment,
what is the next step to driver? (6) Where should we set up the Electric station?
In this electricity replenishing interaction process (figure 11), there are three
agents: Electricity stations, Control Center and Electric-taxis. The detail of process
is described as below: In the beginning user execute the simulation (step1).
Immediately, simulation initial and Electricity stations, Control Center and Electric-
taxis (step2-2, 2-3 and 2-4), and simulation time is beginning running and
updating (step2-1). During the simulation period, Electricity stations will update its’
state (busy/idle) (step3). After generating and simulating for a while, Electric-taxis
can be divided into two kinds of Electric-taxis. Some Electric-taxis are in
replenishing and others are still in operation with electricity shortage (represented
by the longest transverse thick-line). Electric-taxis those are in operation with
electricity shortage ask electricity shortage information to Control Center and other
Electric-taxis those are in replenishing (step4-1 and 4-2). The Electric-taxis that
are in replenishing and Control Center to other Electric-taxis those are in operation
(step5-1 and 5-2).
doi:10.3926/jiem.2009.v2n2.p360-386 ©© JIEM, 2009 – 2(2): 360-386 - ISSN: 2013-0953
Vehicle-based interactive management with multi-agent approach 375
Y. M. Chen; B.-Y. Wang
Figure 11. “Electricity replenishing interaction process”.
And then Electric-taxis those are in lack of electricity select a most suitable
Electricity station according evaluation criteria (step6). After, Electric-taxi moves to
selected elected station (step7). When Electric-taxis enter in or leave the selected
Electricity station, the Electricity station will update its’ state (busy/idle) (step8-1
and 10-1). When Electric-taxi arrives at Electricity station, driver has to waiting for
replenishing until there are no cars in front of him (step8-2 and 9). Finally, after
replenishing finished, Electric-taxi will move to the closest stop and stand by for
assignment (step10-2).
3.3 The specification of management policies
In this sub-section, we specify the all of the management policies that deal with
the problems of three scenarios. The detail of all of the management policies is
described as follows.
Dial-a-ride management
How to choose the proper car to pick Passenger up and deliver:
There are eight stops in our system. Every Electric-taxi has to stand by in
the stop, so Control Center choose the Electric-taxi which is idle and the
distance from the Electric-taxi to specific Passenger is the shortest. The
calculation of distance between Electric-taxi and Passenger is just calculated
doi:10.3926/jiem.2009.v2n2.p360-386 ©© JIEM, 2009 – 2(2): 360-386 - ISSN: 2013-0953
Vehicle-based interactive management with multi-agent approach 376
Y. M. Chen; B.-Y. Wang
based on the distance of the straight line that from Electric-taxi location to
Passenger location.
How to choose the Passenger which you want to pick-up/service:.
When Passengers appear in the system, Control Center will pick the
Passenger who is first entering in the system to service (FIFO (first in first
out)). Then Control Center will chose the second one and so on.
What is the next step to driver, after driver delivering Passenger to his/her
destination: Stand by and check whether Electric-taxi needs to be
replenished or not.
How to choose the pathway to deliver passenger to his/her
destination: Search the shortest path by Dijkstra’s algorithm(Soltani et al.,
2002).
Incident (traffic jam) management
How to know the specific road which is under traffic jam in any
time: If an Electric-taxi get in a road and the car number is more than four,
we treat the state of the Electric-taxi as getting in traffic jam.
How to avoid the specific road which is under traffic jam keep and
moving on: When an Electric-taxi gets in the traffic jam, the Electric-taxi
will send the related traffic jam information (for example: Which road is
under traffic jam) to every Electric-taxi. If other Electric-taxis are going to
enter this road, they will switch into other paths based on the Dijkstra’s
algorithm that excludes the road which is under traffic jam.
Replenishing electricity management
How to schedule Electric-taxis whom are waiting for replenishing:
The Electric-taxi which is first enter in Electric station, he should be served
first (FIFO (first in first out)).
When an Electric-taxi should be replenished: The replenishing
electricity default is 3.8kwh (when the electricity level less than 3.8 so that
can’t afford any longer traveling, the electric-taxi should be replenished).
doi:10.3926/jiem.2009.v2n2.p360-386 ©© JIEM, 2009 – 2(2): 360-386 - ISSN: 2013-0953
Vehicle-based interactive management with multi-agent approach 377
Y. M. Chen; B.-Y. Wang
How long should an Electric-taxi replenish its’ electricity: When the
electricity is up to maximum (19.8kwh), the replenishing stop.
How to select a proper Electric station: The Electric station which is the
closest station and its’ waiting car less than 4 unit. If the closest station is
too busy, we will choose another one which is closer and its’ waiting car less
than 4 unit. “The closest station” means the distance of straight line
between Electric-taxi and Electric station is the shortest.
After finishing the replenishment, what is the next step to driver: Go
back to the closest stop and stand by. “The closest stop” means the
distance of straight line between Electric-taxi and stop is the shortest.
Where should we set up the Electric station: In our system, we apply
random generation to set up Electric station randomly. The default of
Electric station is eight.
In addition, all of the problem about finding out the shortest path, we apply the
Dijkstra’s algorithm to solve it. Following, we specify the algorithm that we use to
solve the shortest path problem.
The description of Dijkstra’s algorithm
Sign Definition
s Passenger start node e Passenger end node N Total node number
Dsj The distance from start node to node j. j = 1~N. Dsj >= 0 (if start node doesn’t connect with node j, then Dsj = 0)
SDj The shortest distances that from start node to node j. j=1~N SDj >= 0 (if start node equals to node j, then SDj = 0)
Cj Whether the node j be chosen or not. j=1~N If Chosen, Cj =1, Otherwise, Cj =0
PNj The pre-node of node j, j=1~N, PNj = 1~N PSN Pre-start node of node j, j=1~N, PSNj = 1~N CD The current shortest distance CN The node of current shortest distance
Table 2. “The signs and their definition of Dijkstra’s algorithm”.
For solving the shortest path problem, the Dijkstra’s algorithm is the most common
algorithm. The objective of this algorithm is to find out the shortest path from start
node to destination node. The important attribute of this algorithm is that we can
find out the shortest path that from start node to each node during the solving.
doi:10.3926/jiem.2009.v2n2.p360-386 ©© JIEM, 2009 – 2(2): 360-386 - ISSN: 2013-0953
Vehicle-based interactive management with multi-agent approach 378
Y. M. Chen; B.-Y. Wang
The following, we list the signs and their definition of Dijkstra’s algorithm, the
detail is described as table 2. Next, we describe the steps of Dijkstra’s algorithm.
In addition, the derivation of Dijkstra’s algorithm can be referred to (Cantone &
Faro, 2004; Solka et al., 1995; Soltani et al., 2002).
Steps of Dijkstra’s algorithm
Step 1. Decide a start node and end node.
Step 2. We set the node of current shortest distance node equals start
node.
Step 3. Setting the distances those from each node to its neighbor nodes.
Step 4. Calculate the distances those from each node to its neighbor nodes.
Step 5. Finding out all of the neighbor nodes to the node of current shortest
distance.
Step 6. Choosing the shortest distance that from the start node to the
specific neighbor node.
Step 7. Setting the neighbor node as the node of current shortest distance.
Step 8. Repeating step 4 to step 7 until all of the nodes has been selected.
Step 9. Obtaining the shortest path that from start node to each node.
Figure 12. “Example of the shortest path problem”.
After understanding the solution steps of Dijkstra’s algorithm, through figure 12 we
can understand the solution procedure more clearly. Figure 12 is a shortest path
doi:10.3926/jiem.2009.v2n2.p360-386 ©© JIEM, 2009 – 2(2): 360-386 - ISSN: 2013-0953
Vehicle-based interactive management with multi-agent approach 379
Y. M. Chen; B.-Y. Wang
problem, our start node is “S” and end node “T”. We want to find out the shortest
path from “S” to ”T”. Following is the solution procedures and description.
Step 1. We set the Start node (s) is S and the end node (e) is T.
Step 2. We set the node of current shortest distance node (CN) is s.
Step 3. All of the distances that from each node to its neighbor nodes are
set as Figure 12.
Step 4. If ( Dsj != 0 || s = = j), then SDj = Dsj; Else, SDj = ∞;
Step 5. For (int j=1; j<=N; j++) {If (SDj < CD) then, CD = SDj; CN = j;
PSN = s; } CCN = 1; For (int j=1; j<N; j++){
Step 6. s = CN for (int j=1; j<N; j++) { If(Dsj + SDs < SDj) then{ SDj =
Dsj + SDs; PNj = s;}}
Step 7. For (int j=1; j<=N; j++) {If (SDj < CD && Cj = = 0) then {CD =
SDj; CN = j; PSN = s; } } CCN = 1; }
Step 8. Repeating step 4 to step 7 until all of the nodes has been selected.
Step 9. Obtaining the shortest path that from S to each node. The shortest
path from “S” to “T” (ScedT, the total distance is 6+2+1+5 = 14) is
represented by thick lines in figure 12.
Following, our creation of simulation system will be present in the Section 4.
4 Creation and simulation of electric-taxi DAR operation system
4.1 Creation of electric-taxi DAR operation system
In our simulation, we use java language as our program language. On the other
hand, AnyLogic is a java-based simulation platform, it just match our need. So we
chose Anylogic as our simulation platform.
doi:10.3926/jiem.2009.v2n2.p360-386 ©© JIEM, 2009 – 2(2): 360-386 - ISSN: 2013-0953
Vehicle-based interactive management with multi-agent approach 380
Y. M. Chen; B.-Y. Wang
Figure 13(a). “The agent attributes of electric-taxi DAR operation system”.
Figure 13(b). “The agent state of electric-taxi DAR operation system”.
According to the framework that proposed in the Section 2 and 3, we have
constructed the multi-agent electric-taxi DAR operation system (figure 14). Figure
13(a)~13(c) are the decomposition of the multi-agent electric-taxi DAR operation
system. Figure 13(a) illustrates the attributes of agent. The attributes of agent are
extracted from the framework of agent. Figure 13(b) illustrates the
doi:10.3926/jiem.2009.v2n2.p360-386 ©© JIEM, 2009 – 2(2): 360-386 - ISSN: 2013-0953
Vehicle-based interactive management with multi-agent approach 381
Y. M. Chen; B.-Y. Wang
states/behaviors of agent. The behaviors of agent are extracted from combining
the framework of agent and interaction process. Figure 13(c) illustrates the
management policies of electric-taxi DAR operation system. The management
policies are completed by specific definition in the Section 3.
Figure 13(c). “The management policies of electric-taxi DAR operation system”.
Figure 14. “Electric-taxi DAR operation simulation system”.
doi:10.3926/jiem.2009.v2n2.p360-386 ©© JIEM, 2009 – 2(2): 360-386 - ISSN: 2013-0953
Vehicle-based interactive management with multi-agent approach 382
Y. M. Chen; B.-Y. Wang
4.2 Simulation of electric-taxi DAR operation system and setting
In this sub-section, we list the environment specification and simulation settings in
the table 3 and table 4. The figure 15 is the simulation of electric-taxi DAR
operation system.
Figure 15. “The simulation of electric-taxi DAR operation system”.
In computer environment, we use Microsoft Windows XP Professional version 2002
Service pack 3 as our operation system. The CPU computer is Intel(R) Core(TM)2
6320 @ 1.86GHz and the RAM computer is 1.87Ghz, 0.99GB. The AnyLogic version
is anylogic 5.1 and the java version is JDK6.0.
In simulation settings, the number of taxi is 10. The number of stop and road is 8
and 18, respectively. And stop generation belongs to random generation. Other
detail information is listed in table 4.
Operation system CPU Ram Anylogic version Java version
Microsoft Windows XP Professional version 2002 Service pack 3
Intel(R) Core(TM)2 6320 @ 1.86GHz
1.87Ghz, 0.99GB
Anylogic 5.1 JDK 6.0
Table 3. “Environment specification”.
Time accuracy
Stop at time
Simulation speed
Model time units per second
The number of taxi
The generation rate of passenger
0.01 30 units Virtual mode 100 units 10 Normal (0.2)
Table 4. “Simulation settings in Anylogic 5.1 version”.
doi:10.3926/jiem.2009.v2n2.p360-386 ©© JIEM, 2009 – 2(2): 360-386 - ISSN: 2013-0953
Vehicle-based interactive management with multi-agent approach 383
Y. M. Chen; B.-Y. Wang
No. Initial parameter / Performance measurement Value
1 Passenger arrival time Normal (0.2) 2 Electric car speed 100 km/ per time unit 3 Electricity 19.8kwh 4 Electricity consuming rate 0.035kwh/ per time unit 5 Average waiting time
Table 5. “Initial parameters and Performance measurement in the electric taxi DAR operation
system”.
We list the initial parameters and performance measurements in the electric taxi
DAR operation system in table 5. The underlying contribution of first four
parameters is to help the simulation approach reality and they are the necessary
simulation parameters for electric taxi DAR operation system. On the other hand,
the underlying contribution of last performance measure (Average waiting time) is
to assess the result of electric taxi DAR operation system simulation.
5 Data collection and analysis
Data generated from the simulation that combines three scenarios (DAR, incident
and replenishing electricity) and management policies specified in 3.2 and 3.3
sections.
From figure 16, the left bar represents the total serviced passenger number and
the number is 152. The middle bar represents average waiting time of passengers
and the time is 3.93. The right bar represents the total passenger number in
simulation system and the number is 410. So we can find out the average waiting
time is just 3.93. That means each of passenger can be served very quickly. The
main cause of this phenomenon is resulted from the average waiting time is only
calculated by the passengers whom has satisfied their demand. There are many
passengers wait in the system, but not be included in the calculation.
On the other hand, the service rate is only (152/410)*100% = 37.073%. That
means the service rate is very low. The main cause is that the generation rate of
passengers is too high, so all of the electric-taxis can’t cope with the large
demands.
doi:10.3926/jiem.2009.v2n2.p360-386 ©© JIEM, 2009 – 2(2): 360-386 - ISSN: 2013-0953
Vehicle-based interactive management with multi-agent approach 384
Y. M. Chen; B.-Y. Wang
Figure 16. “View of a simulation”.
6 Summarizes and future works
We propose an electric-taxi DAR operation system that improves the lacks of
existing methods. And from the analysis result, we successfully obtain the
performance measures (average waiting time and service rate) to support the
decision making with management policies.
In the future works, first, we will design different management policies combine
different scenarios to observe the result of simulation and analyze the
phenomenon. Second, we will construct graphic user interface (GUI) to connect
with on-line. In this way, we can modify our electric-taxi DAR operation system to
approach reality. For electric-taxi DAR managers, they can use our simulation
system as the support of making decision.
Acknowledgements
This research work was sponsored by the National Science Council, R.O.C., under
project number NSC97-2221-E-155-039.
doi:10.3926/jiem.2009.v2n2.p360-386 ©© JIEM, 2009 – 2(2): 360-386 - ISSN: 2013-0953
Vehicle-based interactive management with multi-agent approach 385
Y. M. Chen; B.-Y. Wang
References
Ali, W. (2006). Developping 2D and 3D multi-agent geosimulation, a method and
and its’ application: the case of shopping behavior geosimulation in square one
mall. Retrieved 2006, from www.theses.ulaval.ca/2006/23343
BigSolar (2005). General report. Retrieved 2005, from
http://www.bigsolar.url.tw/F6DM.htm
Cantone, D., & Faro, S. (2004). Two-Levels-Greedy: a generalization of Dijkstra's
shortest path algorithm. Electronic Notes in Discrete Mathematics, 17, 81-86.
Cubillos, C., Polanco, F. G., & Demartini, C. (2008a). Passengers Trips Planning
using Contract-Net with Filters. Paper presented at the 8th International IEEE
Conference on Intelligent Transportation Systems Vienna, Austria.
Ezzedine, H., Kolski, C., & Peninou, A. (2005). Agent-oriented design of human-
computer interface: application to supervision of an urban transport network.
Engineering Applications of Artificial Intelligence, 18, 255–270.
Galus, M. et al. (2009). A Framework for Investigating the Impact of PHEVS.
Retrieved April 7th, 2009, from http://www.eeh.ee.ethz.ch/en/eeh/about-us.html
Government of Honk Kong (2005) General report. Retrieved 2005 from
http://www.epd.gov.hk/epd/partnership/chi/tran.htm
GreenParttyTaiwan (2007). General report, Retrieved 2007 from
http://www.greenparty.org.tw/division.php?itemid=831
iCo2l (2008). General report. Retrieved 2008 from
http://icool.saveoursky.org.tw/co2news/index.php?load=read&id=399
Kok, I. D., & Lucassen, T. (2007). Using Sectors in a Multi Agent Approach to a Taxi
Planning Problem. Retrieved 2009 from http://www.teunlucassen.nl/index.php
Lansdowne, A. (2006). Traffic Simulation using Agent Based Modeling. Retrieved
2009, from http://www.myhomezone.co.uk/project/
doi:10.3926/jiem.2009.v2n2.p360-386 ©© JIEM, 2009 – 2(2): 360-386 - ISSN: 2013-0953
Vehicle-based interactive management with multi-agent approach 386
Y. M. Chen; B.-Y. Wang
Seow, K. T., Dang, N. H., & Lee, D. H. (2008). Using Intelligent Collaborative
Agents for Automating Distributed Taxi Dispatch. Retrieved 2008, from
http://cts.cs.uic.edu/
Siebers, P.O., Aickelin, U., Celia, H., & Clegg, C. (2008). Using Multi-Agent
Simulation to Understand the Impact of Management Practices on Retail
Performance. Retrieved 2008, from http://www.nottingham.ac.uk/cs/
Solka, J. L., Perry, J. C., Poellinger, B. R., & Rogers, G. W. (1995). Fast
computation of optimal paths using a parallel Dijkstra algorithm with embedded
constraints. Neurocomputing, 8(2), 195-212.
Soltani, A. R., Tawfik, H., Goulermas, J. Y., & Fernando, T. (2002). Path planning in
construction sites: performance evaluation of the Dijkstra, A*, and GA search
algorithms. Advanced Engineering Informatics, 16(4), 291-303.
Taiwan Environmental Information Center (2009). General report. Retrieved 2009,
from http://e-info.org.tw/taxonomy/term/122
Wu, C. C. (2006). Matching Models and Solution Algorithms for Urban Taxipool.
Retrieved 2006, from from http://etds.ncl.edu.tw/
Wu, Y. W. (2005). A Study of Shuttle Bus with Dial-a-Ride Service for Long-
Distance Transportation Terminals. Retrieved 2005 from
http://etds.ncl.edu.tw/theabs/index.jsp
©© Journal of Industrial Engineering and Management, 2009 (www.jiem.org)
Article's contents are provided on a Attribution-Non Commercial 3.0 Creative commons license. Readers are allowed to copy, distribute and communicate article's contents, provided the author's and Journal of Industrial
Engineering and Management's names are included. It must not be used for commercial purposes. To see the complete license contents, please visit http://creativecommons.org/licenses/by-nc/3.0/.